blob: 07fb2dfaae13f02588128ea0d73372a8859b63ba [file] [log] [blame]
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001/*
2 * NVM Express device driver
3 * Copyright (c) 2011, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19#include <linux/nvme.h>
20#include <linux/bio.h>
Matthew Wilcox8de05532011-05-12 13:50:28 -040021#include <linux/bitops.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050022#include <linux/blkdev.h>
Matthew Wilcoxfd63e9ce2011-05-06 08:37:54 -040023#include <linux/delay.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050024#include <linux/errno.h>
25#include <linux/fs.h>
26#include <linux/genhd.h>
Matthew Wilcox5aff9382011-05-06 08:45:47 -040027#include <linux/idr.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050028#include <linux/init.h>
29#include <linux/interrupt.h>
30#include <linux/io.h>
31#include <linux/kdev_t.h>
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -050032#include <linux/kthread.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050033#include <linux/kernel.h>
34#include <linux/mm.h>
35#include <linux/module.h>
36#include <linux/moduleparam.h>
37#include <linux/pci.h>
Matthew Wilcoxbe7b6272011-02-06 07:53:23 -050038#include <linux/poison.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050039#include <linux/sched.h>
40#include <linux/slab.h>
41#include <linux/types.h>
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050042
Hitoshi Mitake797a7962012-02-07 11:45:33 +090043#include <asm-generic/io-64-nonatomic-lo-hi.h>
44
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050045#define NVME_Q_DEPTH 1024
46#define SQ_SIZE(depth) (depth * sizeof(struct nvme_command))
47#define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion))
48#define NVME_MINORS 64
Matthew Wilcoxff976d72011-12-20 13:53:01 -050049#define NVME_IO_TIMEOUT (5 * HZ)
Matthew Wilcoxe85248e2011-02-06 18:30:16 -050050#define ADMIN_TIMEOUT (60 * HZ)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050051
52static int nvme_major;
53module_param(nvme_major, int, 0);
54
Matthew Wilcox58ffacb2011-02-06 07:28:06 -050055static int use_threaded_interrupts;
56module_param(use_threaded_interrupts, int, 0);
57
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -050058static DEFINE_SPINLOCK(dev_list_lock);
59static LIST_HEAD(dev_list);
60static struct task_struct *nvme_thread;
61
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050062/*
63 * Represents an NVM Express device. Each nvme_dev is a PCI function.
64 */
65struct nvme_dev {
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -050066 struct list_head node;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050067 struct nvme_queue **queues;
68 u32 __iomem *dbs;
69 struct pci_dev *pci_dev;
Matthew Wilcox091b6092011-02-10 09:56:01 -050070 struct dma_pool *prp_page_pool;
Matthew Wilcox99802a72011-02-10 10:30:34 -050071 struct dma_pool *prp_small_pool;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050072 int instance;
73 int queue_count;
Matthew Wilcoxf1938f62011-10-20 17:00:41 -040074 int db_stride;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050075 u32 ctrl_config;
76 struct msix_entry *entry;
77 struct nvme_bar __iomem *bar;
78 struct list_head namespaces;
Matthew Wilcox51814232011-02-01 16:18:08 -050079 char serial[20];
80 char model[40];
81 char firmware_rev[8];
Keith Busch8fc23e02012-07-26 11:29:57 -060082 u32 max_hw_sectors;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -050083};
84
85/*
86 * An NVM Express namespace is equivalent to a SCSI LUN
87 */
88struct nvme_ns {
89 struct list_head list;
90
91 struct nvme_dev *dev;
92 struct request_queue *queue;
93 struct gendisk *disk;
94
95 int ns_id;
96 int lba_shift;
97};
98
99/*
100 * An NVM Express queue. Each device has at least two (one for admin
101 * commands and one for I/O commands).
102 */
103struct nvme_queue {
104 struct device *q_dmadev;
Matthew Wilcox091b6092011-02-10 09:56:01 -0500105 struct nvme_dev *dev;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500106 spinlock_t q_lock;
107 struct nvme_command *sq_cmds;
108 volatile struct nvme_completion *cqes;
109 dma_addr_t sq_dma_addr;
110 dma_addr_t cq_dma_addr;
111 wait_queue_head_t sq_full;
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -0500112 wait_queue_t sq_cong_wait;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500113 struct bio_list sq_cong;
114 u32 __iomem *q_db;
115 u16 q_depth;
116 u16 cq_vector;
117 u16 sq_head;
118 u16 sq_tail;
119 u16 cq_head;
Matthew Wilcox82123462011-01-20 13:24:06 -0500120 u16 cq_phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500121 unsigned long cmdid_data[];
122};
123
124/*
125 * Check we didin't inadvertently grow the command struct
126 */
127static inline void _nvme_check_size(void)
128{
129 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
130 BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64);
131 BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64);
132 BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
133 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
134 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
135 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
136 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
137 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
138}
139
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500140typedef void (*nvme_completion_fn)(struct nvme_dev *, void *,
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400141 struct nvme_completion *);
142
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500143struct nvme_cmd_info {
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400144 nvme_completion_fn fn;
145 void *ctx;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500146 unsigned long timeout;
147};
148
149static struct nvme_cmd_info *nvme_cmd_info(struct nvme_queue *nvmeq)
150{
151 return (void *)&nvmeq->cmdid_data[BITS_TO_LONGS(nvmeq->q_depth)];
152}
153
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500154/**
Matthew Wilcox714a7a22011-03-16 16:28:24 -0400155 * alloc_cmdid() - Allocate a Command ID
156 * @nvmeq: The queue that will be used for this command
157 * @ctx: A pointer that will be passed to the handler
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400158 * @handler: The function to call on completion
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500159 *
160 * Allocate a Command ID for a queue. The data passed in will
161 * be passed to the completion handler. This is implemented by using
162 * the bottom two bits of the ctx pointer to store the handler ID.
163 * Passing in a pointer that's not 4-byte aligned will cause a BUG.
164 * We can change this if it becomes a problem.
Matthew Wilcox184d2942011-05-11 21:36:38 -0400165 *
166 * May be called with local interrupts disabled and the q_lock held,
167 * or with interrupts enabled and no locks held.
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500168 */
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400169static int alloc_cmdid(struct nvme_queue *nvmeq, void *ctx,
170 nvme_completion_fn handler, unsigned timeout)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500171{
Matthew Wilcoxe6d15f72011-02-24 08:49:41 -0500172 int depth = nvmeq->q_depth - 1;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500173 struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500174 int cmdid;
175
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500176 do {
177 cmdid = find_first_zero_bit(nvmeq->cmdid_data, depth);
178 if (cmdid >= depth)
179 return -EBUSY;
180 } while (test_and_set_bit(cmdid, nvmeq->cmdid_data));
181
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400182 info[cmdid].fn = handler;
183 info[cmdid].ctx = ctx;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500184 info[cmdid].timeout = jiffies + timeout;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500185 return cmdid;
186}
187
188static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx,
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400189 nvme_completion_fn handler, unsigned timeout)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500190{
191 int cmdid;
192 wait_event_killable(nvmeq->sq_full,
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500193 (cmdid = alloc_cmdid(nvmeq, ctx, handler, timeout)) >= 0);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500194 return (cmdid < 0) ? -EINTR : cmdid;
195}
196
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400197/* Special values must be less than 0x1000 */
198#define CMD_CTX_BASE ((void *)POISON_POINTER_DELTA)
Matthew Wilcoxd2d87032011-02-07 15:55:59 -0500199#define CMD_CTX_CANCELLED (0x30C + CMD_CTX_BASE)
200#define CMD_CTX_COMPLETED (0x310 + CMD_CTX_BASE)
201#define CMD_CTX_INVALID (0x314 + CMD_CTX_BASE)
Matthew Wilcox00df5cb2011-02-22 14:18:30 -0500202#define CMD_CTX_FLUSH (0x318 + CMD_CTX_BASE)
Matthew Wilcoxbe7b6272011-02-06 07:53:23 -0500203
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500204static void special_completion(struct nvme_dev *dev, void *ctx,
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400205 struct nvme_completion *cqe)
206{
207 if (ctx == CMD_CTX_CANCELLED)
208 return;
209 if (ctx == CMD_CTX_FLUSH)
210 return;
211 if (ctx == CMD_CTX_COMPLETED) {
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500212 dev_warn(&dev->pci_dev->dev,
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400213 "completed id %d twice on queue %d\n",
214 cqe->command_id, le16_to_cpup(&cqe->sq_id));
215 return;
216 }
217 if (ctx == CMD_CTX_INVALID) {
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500218 dev_warn(&dev->pci_dev->dev,
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400219 "invalid id %d completed on queue %d\n",
220 cqe->command_id, le16_to_cpup(&cqe->sq_id));
221 return;
222 }
223
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500224 dev_warn(&dev->pci_dev->dev, "Unknown special completion %p\n", ctx);
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400225}
226
Matthew Wilcox184d2942011-05-11 21:36:38 -0400227/*
228 * Called with local interrupts disabled and the q_lock held. May not sleep.
229 */
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400230static void *free_cmdid(struct nvme_queue *nvmeq, int cmdid,
231 nvme_completion_fn *fn)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500232{
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400233 void *ctx;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500234 struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500235
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400236 if (cmdid >= nvmeq->q_depth) {
237 *fn = special_completion;
Matthew Wilcox48e3d392011-02-06 08:51:15 -0500238 return CMD_CTX_INVALID;
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400239 }
240 *fn = info[cmdid].fn;
241 ctx = info[cmdid].ctx;
242 info[cmdid].fn = special_completion;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500243 info[cmdid].ctx = CMD_CTX_COMPLETED;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500244 clear_bit(cmdid, nvmeq->cmdid_data);
245 wake_up(&nvmeq->sq_full);
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400246 return ctx;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500247}
248
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400249static void *cancel_cmdid(struct nvme_queue *nvmeq, int cmdid,
250 nvme_completion_fn *fn)
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500251{
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400252 void *ctx;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500253 struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400254 if (fn)
255 *fn = info[cmdid].fn;
256 ctx = info[cmdid].ctx;
257 info[cmdid].fn = special_completion;
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500258 info[cmdid].ctx = CMD_CTX_CANCELLED;
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400259 return ctx;
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500260}
261
Matthew Wilcox040a93b2011-12-20 11:04:12 -0500262static struct nvme_queue *get_nvmeq(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500263{
Matthew Wilcox040a93b2011-12-20 11:04:12 -0500264 return dev->queues[get_cpu() + 1];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500265}
266
267static void put_nvmeq(struct nvme_queue *nvmeq)
268{
Matthew Wilcox1b234842011-01-20 13:01:49 -0500269 put_cpu();
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500270}
271
272/**
Matthew Wilcox714a7a22011-03-16 16:28:24 -0400273 * nvme_submit_cmd() - Copy a command into a queue and ring the doorbell
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500274 * @nvmeq: The queue to use
275 * @cmd: The command to send
276 *
277 * Safe to use from interrupt context
278 */
279static int nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
280{
281 unsigned long flags;
282 u16 tail;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500283 spin_lock_irqsave(&nvmeq->q_lock, flags);
284 tail = nvmeq->sq_tail;
285 memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500286 if (++tail == nvmeq->q_depth)
287 tail = 0;
Matthew Wilcox75478812011-02-16 09:59:59 -0500288 writel(tail, nvmeq->q_db);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500289 nvmeq->sq_tail = tail;
290 spin_unlock_irqrestore(&nvmeq->q_lock, flags);
291
292 return 0;
293}
294
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500295/*
296 * The nvme_iod describes the data in an I/O, including the list of PRP
297 * entries. You can't see it in this data structure because C doesn't let
298 * me express that. Use nvme_alloc_iod to ensure there's enough space
299 * allocated to store the PRP list.
300 */
301struct nvme_iod {
302 void *private; /* For the use of the submitter of the I/O */
303 int npages; /* In the PRP list. 0 means small pool in use */
304 int offset; /* Of PRP list */
305 int nents; /* Used in scatterlist */
306 int length; /* Of data, in bytes */
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500307 dma_addr_t first_dma;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500308 struct scatterlist sg[0];
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500309};
310
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500311static __le64 **iod_list(struct nvme_iod *iod)
312{
313 return ((void *)iod) + iod->offset;
314}
315
316/*
317 * Will slightly overestimate the number of pages needed. This is OK
318 * as it only leads to a small amount of wasted memory for the lifetime of
319 * the I/O.
320 */
321static int nvme_npages(unsigned size)
322{
323 unsigned nprps = DIV_ROUND_UP(size + PAGE_SIZE, PAGE_SIZE);
324 return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8);
325}
326
327static struct nvme_iod *
328nvme_alloc_iod(unsigned nseg, unsigned nbytes, gfp_t gfp)
329{
330 struct nvme_iod *iod = kmalloc(sizeof(struct nvme_iod) +
331 sizeof(__le64 *) * nvme_npages(nbytes) +
332 sizeof(struct scatterlist) * nseg, gfp);
333
334 if (iod) {
335 iod->offset = offsetof(struct nvme_iod, sg[nseg]);
336 iod->npages = -1;
337 iod->length = nbytes;
338 }
339
340 return iod;
341}
342
343static void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod)
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500344{
345 const int last_prp = PAGE_SIZE / 8 - 1;
346 int i;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500347 __le64 **list = iod_list(iod);
348 dma_addr_t prp_dma = iod->first_dma;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500349
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500350 if (iod->npages == 0)
351 dma_pool_free(dev->prp_small_pool, list[0], prp_dma);
352 for (i = 0; i < iod->npages; i++) {
353 __le64 *prp_list = list[i];
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500354 dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]);
Matthew Wilcox091b6092011-02-10 09:56:01 -0500355 dma_pool_free(dev->prp_page_pool, prp_list, prp_dma);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500356 prp_dma = next_prp_dma;
357 }
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500358 kfree(iod);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500359}
360
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500361static void requeue_bio(struct nvme_dev *dev, struct bio *bio)
362{
363 struct nvme_queue *nvmeq = get_nvmeq(dev);
364 if (bio_list_empty(&nvmeq->sq_cong))
365 add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait);
366 bio_list_add(&nvmeq->sq_cong, bio);
367 put_nvmeq(nvmeq);
368 wake_up_process(nvme_thread);
369}
370
371static void bio_completion(struct nvme_dev *dev, void *ctx,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500372 struct nvme_completion *cqe)
373{
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500374 struct nvme_iod *iod = ctx;
375 struct bio *bio = iod->private;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500376 u16 status = le16_to_cpup(&cqe->status) >> 1;
377
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500378 dma_unmap_sg(&dev->pci_dev->dev, iod->sg, iod->nents,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500379 bio_data_dir(bio) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500380 nvme_free_iod(dev, iod);
Matthew Wilcox09a58f52011-04-28 23:09:09 -0700381 if (status) {
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500382 bio_endio(bio, -EIO);
Matthew Wilcox09a58f52011-04-28 23:09:09 -0700383 } else if (bio->bi_vcnt > bio->bi_idx) {
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500384 requeue_bio(dev, bio);
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500385 } else {
386 bio_endio(bio, 0);
387 }
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500388}
389
Matthew Wilcox184d2942011-05-11 21:36:38 -0400390/* length is in bytes. gfp flags indicates whether we may sleep. */
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500391static int nvme_setup_prps(struct nvme_dev *dev,
392 struct nvme_common_command *cmd, struct nvme_iod *iod,
393 int total_len, gfp_t gfp)
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500394{
Matthew Wilcox99802a72011-02-10 10:30:34 -0500395 struct dma_pool *pool;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500396 int length = total_len;
397 struct scatterlist *sg = iod->sg;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500398 int dma_len = sg_dma_len(sg);
399 u64 dma_addr = sg_dma_address(sg);
400 int offset = offset_in_page(dma_addr);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500401 __le64 *prp_list;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500402 __le64 **list = iod_list(iod);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500403 dma_addr_t prp_dma;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500404 int nprps, i;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500405
406 cmd->prp1 = cpu_to_le64(dma_addr);
407 length -= (PAGE_SIZE - offset);
408 if (length <= 0)
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500409 return total_len;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500410
411 dma_len -= (PAGE_SIZE - offset);
412 if (dma_len) {
413 dma_addr += (PAGE_SIZE - offset);
414 } else {
415 sg = sg_next(sg);
416 dma_addr = sg_dma_address(sg);
417 dma_len = sg_dma_len(sg);
418 }
419
420 if (length <= PAGE_SIZE) {
421 cmd->prp2 = cpu_to_le64(dma_addr);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500422 return total_len;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500423 }
424
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500425 nprps = DIV_ROUND_UP(length, PAGE_SIZE);
Matthew Wilcox99802a72011-02-10 10:30:34 -0500426 if (nprps <= (256 / 8)) {
427 pool = dev->prp_small_pool;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500428 iod->npages = 0;
Matthew Wilcox99802a72011-02-10 10:30:34 -0500429 } else {
430 pool = dev->prp_page_pool;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500431 iod->npages = 1;
Matthew Wilcox99802a72011-02-10 10:30:34 -0500432 }
433
Matthew Wilcoxb77954c2011-05-12 13:51:41 -0400434 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
435 if (!prp_list) {
436 cmd->prp2 = cpu_to_le64(dma_addr);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500437 iod->npages = -1;
438 return (total_len - length) + PAGE_SIZE;
Matthew Wilcoxb77954c2011-05-12 13:51:41 -0400439 }
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500440 list[0] = prp_list;
441 iod->first_dma = prp_dma;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500442 cmd->prp2 = cpu_to_le64(prp_dma);
443 i = 0;
444 for (;;) {
Matthew Wilcox7523d832011-03-16 16:43:40 -0400445 if (i == PAGE_SIZE / 8) {
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500446 __le64 *old_prp_list = prp_list;
Matthew Wilcoxb77954c2011-05-12 13:51:41 -0400447 prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500448 if (!prp_list)
449 return total_len - length;
450 list[iod->npages++] = prp_list;
Matthew Wilcox7523d832011-03-16 16:43:40 -0400451 prp_list[0] = old_prp_list[i - 1];
452 old_prp_list[i - 1] = cpu_to_le64(prp_dma);
453 i = 1;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -0500454 }
455 prp_list[i++] = cpu_to_le64(dma_addr);
456 dma_len -= PAGE_SIZE;
457 dma_addr += PAGE_SIZE;
458 length -= PAGE_SIZE;
459 if (length <= 0)
460 break;
461 if (dma_len > 0)
462 continue;
463 BUG_ON(dma_len < 0);
464 sg = sg_next(sg);
465 dma_addr = sg_dma_address(sg);
466 dma_len = sg_dma_len(sg);
467 }
468
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500469 return total_len;
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500470}
471
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500472/* NVMe scatterlists require no holes in the virtual address */
473#define BIOVEC_NOT_VIRT_MERGEABLE(vec1, vec2) ((vec2)->bv_offset || \
474 (((vec1)->bv_offset + (vec1)->bv_len) % PAGE_SIZE))
475
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500476static int nvme_map_bio(struct device *dev, struct nvme_iod *iod,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500477 struct bio *bio, enum dma_data_direction dma_dir, int psegs)
478{
Matthew Wilcox76830842011-02-10 13:55:39 -0500479 struct bio_vec *bvec, *bvprv = NULL;
480 struct scatterlist *sg = NULL;
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500481 int i, old_idx, length = 0, nsegs = 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500482
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500483 sg_init_table(iod->sg, psegs);
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500484 old_idx = bio->bi_idx;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500485 bio_for_each_segment(bvec, bio, i) {
Matthew Wilcox76830842011-02-10 13:55:39 -0500486 if (bvprv && BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) {
487 sg->length += bvec->bv_len;
488 } else {
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500489 if (bvprv && BIOVEC_NOT_VIRT_MERGEABLE(bvprv, bvec))
490 break;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500491 sg = sg ? sg + 1 : iod->sg;
Matthew Wilcox76830842011-02-10 13:55:39 -0500492 sg_set_page(sg, bvec->bv_page, bvec->bv_len,
493 bvec->bv_offset);
494 nsegs++;
495 }
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500496 length += bvec->bv_len;
Matthew Wilcox76830842011-02-10 13:55:39 -0500497 bvprv = bvec;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500498 }
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500499 bio->bi_idx = i;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500500 iod->nents = nsegs;
Matthew Wilcox76830842011-02-10 13:55:39 -0500501 sg_mark_end(sg);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500502 if (dma_map_sg(dev, iod->sg, iod->nents, dma_dir) == 0) {
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500503 bio->bi_idx = old_idx;
504 return -ENOMEM;
505 }
506 return length;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500507}
508
Matthew Wilcox00df5cb2011-02-22 14:18:30 -0500509static int nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns,
510 int cmdid)
511{
512 struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail];
513
514 memset(cmnd, 0, sizeof(*cmnd));
515 cmnd->common.opcode = nvme_cmd_flush;
516 cmnd->common.command_id = cmdid;
517 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
518
519 if (++nvmeq->sq_tail == nvmeq->q_depth)
520 nvmeq->sq_tail = 0;
521 writel(nvmeq->sq_tail, nvmeq->q_db);
522
523 return 0;
524}
525
526static int nvme_submit_flush_data(struct nvme_queue *nvmeq, struct nvme_ns *ns)
527{
528 int cmdid = alloc_cmdid(nvmeq, (void *)CMD_CTX_FLUSH,
Matthew Wilcoxff976d72011-12-20 13:53:01 -0500529 special_completion, NVME_IO_TIMEOUT);
Matthew Wilcox00df5cb2011-02-22 14:18:30 -0500530 if (unlikely(cmdid < 0))
531 return cmdid;
532
533 return nvme_submit_flush(nvmeq, ns, cmdid);
534}
535
Matthew Wilcox184d2942011-05-11 21:36:38 -0400536/*
537 * Called with local interrupts disabled and the q_lock held. May not sleep.
538 */
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500539static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns,
540 struct bio *bio)
541{
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500542 struct nvme_command *cmnd;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500543 struct nvme_iod *iod;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500544 enum dma_data_direction dma_dir;
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500545 int cmdid, length, result = -ENOMEM;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500546 u16 control;
547 u32 dsmgmt;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500548 int psegs = bio_phys_segments(ns->queue, bio);
549
Matthew Wilcox00df5cb2011-02-22 14:18:30 -0500550 if ((bio->bi_rw & REQ_FLUSH) && psegs) {
551 result = nvme_submit_flush_data(nvmeq, ns);
552 if (result)
553 return result;
554 }
555
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500556 iod = nvme_alloc_iod(psegs, bio->bi_size, GFP_ATOMIC);
557 if (!iod)
Matthew Wilcoxeeee3222011-02-14 15:55:33 -0500558 goto nomem;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500559 iod->private = bio;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500560
Matthew Wilcoxeeee3222011-02-14 15:55:33 -0500561 result = -EBUSY;
Matthew Wilcoxff976d72011-12-20 13:53:01 -0500562 cmdid = alloc_cmdid(nvmeq, iod, bio_completion, NVME_IO_TIMEOUT);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500563 if (unlikely(cmdid < 0))
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500564 goto free_iod;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500565
Matthew Wilcox00df5cb2011-02-22 14:18:30 -0500566 if ((bio->bi_rw & REQ_FLUSH) && !psegs)
567 return nvme_submit_flush(nvmeq, ns, cmdid);
568
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500569 control = 0;
570 if (bio->bi_rw & REQ_FUA)
571 control |= NVME_RW_FUA;
572 if (bio->bi_rw & (REQ_FAILFAST_DEV | REQ_RAHEAD))
573 control |= NVME_RW_LR;
574
575 dsmgmt = 0;
576 if (bio->bi_rw & REQ_RAHEAD)
577 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
578
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500579 cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500580
Matthew Wilcoxb8deb622011-01-26 10:08:25 -0500581 memset(cmnd, 0, sizeof(*cmnd));
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500582 if (bio_data_dir(bio)) {
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500583 cmnd->rw.opcode = nvme_cmd_write;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500584 dma_dir = DMA_TO_DEVICE;
585 } else {
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500586 cmnd->rw.opcode = nvme_cmd_read;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500587 dma_dir = DMA_FROM_DEVICE;
588 }
589
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500590 result = nvme_map_bio(nvmeq->q_dmadev, iod, bio, dma_dir, psegs);
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500591 if (result < 0)
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500592 goto free_iod;
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500593 length = result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500594
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500595 cmnd->rw.command_id = cmdid;
596 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500597 length = nvme_setup_prps(nvmeq->dev, &cmnd->common, iod, length,
598 GFP_ATOMIC);
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500599 cmnd->rw.slba = cpu_to_le64(bio->bi_sector >> (ns->lba_shift - 9));
Matthew Wilcox1ad2f892011-02-23 15:20:00 -0500600 cmnd->rw.length = cpu_to_le16((length >> ns->lba_shift) - 1);
Matthew Wilcoxff22b542011-01-26 10:02:29 -0500601 cmnd->rw.control = cpu_to_le16(control);
602 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500603
Matthew Wilcoxd8ee9d62011-02-24 08:46:00 -0500604 bio->bi_sector += length >> 9;
605
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500606 if (++nvmeq->sq_tail == nvmeq->q_depth)
607 nvmeq->sq_tail = 0;
Matthew Wilcox75478812011-02-16 09:59:59 -0500608 writel(nvmeq->sq_tail, nvmeq->q_db);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500609
Matthew Wilcox1974b1a2011-02-10 12:01:09 -0500610 return 0;
611
Matthew Wilcoxeca18b22011-12-20 13:34:52 -0500612 free_iod:
613 nvme_free_iod(nvmeq->dev, iod);
Matthew Wilcoxeeee3222011-02-14 15:55:33 -0500614 nomem:
615 return result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500616}
617
Linus Torvalds93c3d652012-01-18 15:41:27 -0800618static void nvme_make_request(struct request_queue *q, struct bio *bio)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500619{
620 struct nvme_ns *ns = q->queuedata;
Matthew Wilcox040a93b2011-12-20 11:04:12 -0500621 struct nvme_queue *nvmeq = get_nvmeq(ns->dev);
Matthew Wilcoxeeee3222011-02-14 15:55:33 -0500622 int result = -EBUSY;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500623
Matthew Wilcoxeeee3222011-02-14 15:55:33 -0500624 spin_lock_irq(&nvmeq->q_lock);
625 if (bio_list_empty(&nvmeq->sq_cong))
626 result = nvme_submit_bio_queue(nvmeq, ns, bio);
627 if (unlikely(result)) {
628 if (bio_list_empty(&nvmeq->sq_cong))
629 add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500630 bio_list_add(&nvmeq->sq_cong, bio);
631 }
Matthew Wilcoxeeee3222011-02-14 15:55:33 -0500632
633 spin_unlock_irq(&nvmeq->q_lock);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500634 put_nvmeq(nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500635}
636
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500637static irqreturn_t nvme_process_cq(struct nvme_queue *nvmeq)
638{
Matthew Wilcox82123462011-01-20 13:24:06 -0500639 u16 head, phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500640
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500641 head = nvmeq->cq_head;
Matthew Wilcox82123462011-01-20 13:24:06 -0500642 phase = nvmeq->cq_phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500643
644 for (;;) {
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400645 void *ctx;
646 nvme_completion_fn fn;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500647 struct nvme_completion cqe = nvmeq->cqes[head];
Matthew Wilcox82123462011-01-20 13:24:06 -0500648 if ((le16_to_cpu(cqe.status) & 1) != phase)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500649 break;
650 nvmeq->sq_head = le16_to_cpu(cqe.sq_head);
651 if (++head == nvmeq->q_depth) {
652 head = 0;
Matthew Wilcox82123462011-01-20 13:24:06 -0500653 phase = !phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500654 }
655
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400656 ctx = free_cmdid(nvmeq, cqe.command_id, &fn);
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500657 fn(nvmeq->dev, ctx, &cqe);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500658 }
659
660 /* If the controller ignores the cq head doorbell and continuously
661 * writes to the queue, it is theoretically possible to wrap around
662 * the queue twice and mistakenly return IRQ_NONE. Linux only
663 * requires that 0.1% of your interrupts are handled, so this isn't
664 * a big problem.
665 */
Matthew Wilcox82123462011-01-20 13:24:06 -0500666 if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500667 return IRQ_NONE;
668
Matthew Wilcoxf1938f62011-10-20 17:00:41 -0400669 writel(head, nvmeq->q_db + (1 << nvmeq->dev->db_stride));
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500670 nvmeq->cq_head = head;
Matthew Wilcox82123462011-01-20 13:24:06 -0500671 nvmeq->cq_phase = phase;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500672
673 return IRQ_HANDLED;
674}
675
676static irqreturn_t nvme_irq(int irq, void *data)
677{
Matthew Wilcox58ffacb2011-02-06 07:28:06 -0500678 irqreturn_t result;
679 struct nvme_queue *nvmeq = data;
680 spin_lock(&nvmeq->q_lock);
681 result = nvme_process_cq(nvmeq);
682 spin_unlock(&nvmeq->q_lock);
683 return result;
684}
685
686static irqreturn_t nvme_irq_check(int irq, void *data)
687{
688 struct nvme_queue *nvmeq = data;
689 struct nvme_completion cqe = nvmeq->cqes[nvmeq->cq_head];
690 if ((le16_to_cpu(cqe.status) & 1) != nvmeq->cq_phase)
691 return IRQ_NONE;
692 return IRQ_WAKE_THREAD;
693}
694
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500695static void nvme_abort_command(struct nvme_queue *nvmeq, int cmdid)
696{
697 spin_lock_irq(&nvmeq->q_lock);
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400698 cancel_cmdid(nvmeq, cmdid, NULL);
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500699 spin_unlock_irq(&nvmeq->q_lock);
700}
701
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400702struct sync_cmd_info {
703 struct task_struct *task;
704 u32 result;
705 int status;
706};
707
Matthew Wilcox5c1281a2011-12-20 11:54:53 -0500708static void sync_completion(struct nvme_dev *dev, void *ctx,
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400709 struct nvme_completion *cqe)
710{
711 struct sync_cmd_info *cmdinfo = ctx;
712 cmdinfo->result = le32_to_cpup(&cqe->result);
713 cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
714 wake_up_process(cmdinfo->task);
715}
716
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500717/*
718 * Returns 0 on success. If the result is negative, it's a Linux error code;
719 * if the result is positive, it's an NVM Express status code
720 */
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500721static int nvme_submit_sync_cmd(struct nvme_queue *nvmeq,
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500722 struct nvme_command *cmd, u32 *result, unsigned timeout)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500723{
724 int cmdid;
725 struct sync_cmd_info cmdinfo;
726
727 cmdinfo.task = current;
728 cmdinfo.status = -EINTR;
729
Matthew Wilcoxc2f5b652011-10-15 07:33:46 -0400730 cmdid = alloc_cmdid_killable(nvmeq, &cmdinfo, sync_completion,
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500731 timeout);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500732 if (cmdid < 0)
733 return cmdid;
734 cmd->common.command_id = cmdid;
735
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500736 set_current_state(TASK_KILLABLE);
737 nvme_submit_cmd(nvmeq, cmd);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500738 schedule();
739
Matthew Wilcox3c0cf132011-02-04 16:03:56 -0500740 if (cmdinfo.status == -EINTR) {
741 nvme_abort_command(nvmeq, cmdid);
742 return -EINTR;
743 }
744
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500745 if (result)
746 *result = cmdinfo.result;
747
748 return cmdinfo.status;
749}
750
751static int nvme_submit_admin_cmd(struct nvme_dev *dev, struct nvme_command *cmd,
752 u32 *result)
753{
Matthew Wilcoxe85248e2011-02-06 18:30:16 -0500754 return nvme_submit_sync_cmd(dev->queues[0], cmd, result, ADMIN_TIMEOUT);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500755}
756
757static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
758{
759 int status;
760 struct nvme_command c;
761
762 memset(&c, 0, sizeof(c));
763 c.delete_queue.opcode = opcode;
764 c.delete_queue.qid = cpu_to_le16(id);
765
766 status = nvme_submit_admin_cmd(dev, &c, NULL);
767 if (status)
768 return -EIO;
769 return 0;
770}
771
772static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid,
773 struct nvme_queue *nvmeq)
774{
775 int status;
776 struct nvme_command c;
777 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED;
778
779 memset(&c, 0, sizeof(c));
780 c.create_cq.opcode = nvme_admin_create_cq;
781 c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr);
782 c.create_cq.cqid = cpu_to_le16(qid);
783 c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
784 c.create_cq.cq_flags = cpu_to_le16(flags);
785 c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector);
786
787 status = nvme_submit_admin_cmd(dev, &c, NULL);
788 if (status)
789 return -EIO;
790 return 0;
791}
792
793static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid,
794 struct nvme_queue *nvmeq)
795{
796 int status;
797 struct nvme_command c;
798 int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM;
799
800 memset(&c, 0, sizeof(c));
801 c.create_sq.opcode = nvme_admin_create_sq;
802 c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr);
803 c.create_sq.sqid = cpu_to_le16(qid);
804 c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1);
805 c.create_sq.sq_flags = cpu_to_le16(flags);
806 c.create_sq.cqid = cpu_to_le16(qid);
807
808 status = nvme_submit_admin_cmd(dev, &c, NULL);
809 if (status)
810 return -EIO;
811 return 0;
812}
813
814static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid)
815{
816 return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid);
817}
818
819static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid)
820{
821 return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid);
822}
823
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400824static int nvme_identify(struct nvme_dev *dev, unsigned nsid, unsigned cns,
825 dma_addr_t dma_addr)
826{
827 struct nvme_command c;
828
829 memset(&c, 0, sizeof(c));
830 c.identify.opcode = nvme_admin_identify;
831 c.identify.nsid = cpu_to_le32(nsid);
832 c.identify.prp1 = cpu_to_le64(dma_addr);
833 c.identify.cns = cpu_to_le32(cns);
834
835 return nvme_submit_admin_cmd(dev, &c, NULL);
836}
837
838static int nvme_get_features(struct nvme_dev *dev, unsigned fid,
Keith Buscha42cecc2012-07-25 16:06:38 -0600839 unsigned nsid, dma_addr_t dma_addr)
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400840{
841 struct nvme_command c;
842
843 memset(&c, 0, sizeof(c));
844 c.features.opcode = nvme_admin_get_features;
Keith Buscha42cecc2012-07-25 16:06:38 -0600845 c.features.nsid = cpu_to_le32(nsid);
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400846 c.features.prp1 = cpu_to_le64(dma_addr);
847 c.features.fid = cpu_to_le32(fid);
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400848
Matthew Wilcoxdf348132012-01-11 07:29:56 -0700849 return nvme_submit_admin_cmd(dev, &c, NULL);
850}
851
852static int nvme_set_features(struct nvme_dev *dev, unsigned fid,
853 unsigned dword11, dma_addr_t dma_addr, u32 *result)
854{
855 struct nvme_command c;
856
857 memset(&c, 0, sizeof(c));
858 c.features.opcode = nvme_admin_set_features;
859 c.features.prp1 = cpu_to_le64(dma_addr);
860 c.features.fid = cpu_to_le32(fid);
861 c.features.dword11 = cpu_to_le32(dword11);
862
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -0400863 return nvme_submit_admin_cmd(dev, &c, result);
864}
865
Matthew Wilcoxa09115b2012-08-07 15:56:23 -0400866/**
867 * nvme_cancel_ios - Cancel outstanding I/Os
868 * @queue: The queue to cancel I/Os on
869 * @timeout: True to only cancel I/Os which have timed out
870 */
871static void nvme_cancel_ios(struct nvme_queue *nvmeq, bool timeout)
872{
873 int depth = nvmeq->q_depth - 1;
874 struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
875 unsigned long now = jiffies;
876 int cmdid;
877
878 for_each_set_bit(cmdid, nvmeq->cmdid_data, depth) {
879 void *ctx;
880 nvme_completion_fn fn;
881 static struct nvme_completion cqe = {
882 .status = cpu_to_le16(NVME_SC_ABORT_REQ) << 1,
883 };
884
885 if (timeout && !time_after(now, info[cmdid].timeout))
886 continue;
887 dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d\n", cmdid);
888 ctx = cancel_cmdid(nvmeq, cmdid, &fn);
889 fn(nvmeq->dev, ctx, &cqe);
890 }
891}
892
Matthew Wilcox9e866772012-08-03 13:55:56 -0400893static void nvme_free_queue_mem(struct nvme_queue *nvmeq)
894{
895 dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
896 (void *)nvmeq->cqes, nvmeq->cq_dma_addr);
897 dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
898 nvmeq->sq_cmds, nvmeq->sq_dma_addr);
899 kfree(nvmeq);
900}
901
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500902static void nvme_free_queue(struct nvme_dev *dev, int qid)
903{
904 struct nvme_queue *nvmeq = dev->queues[qid];
Matthew Wilcoxaba20802011-03-27 08:52:06 -0400905 int vector = dev->entry[nvmeq->cq_vector].vector;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500906
Matthew Wilcoxa09115b2012-08-07 15:56:23 -0400907 spin_lock_irq(&nvmeq->q_lock);
908 nvme_cancel_ios(nvmeq, false);
909 spin_unlock_irq(&nvmeq->q_lock);
910
Matthew Wilcoxaba20802011-03-27 08:52:06 -0400911 irq_set_affinity_hint(vector, NULL);
912 free_irq(vector, nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500913
914 /* Don't tell the adapter to delete the admin queue */
915 if (qid) {
916 adapter_delete_sq(dev, qid);
917 adapter_delete_cq(dev, qid);
918 }
919
Matthew Wilcox9e866772012-08-03 13:55:56 -0400920 nvme_free_queue_mem(nvmeq);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500921}
922
923static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
924 int depth, int vector)
925{
926 struct device *dmadev = &dev->pci_dev->dev;
Keith Buscha0cadb82012-07-27 13:57:23 -0400927 unsigned extra = DIV_ROUND_UP(depth, 8) + (depth *
928 sizeof(struct nvme_cmd_info));
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500929 struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq) + extra, GFP_KERNEL);
930 if (!nvmeq)
931 return NULL;
932
933 nvmeq->cqes = dma_alloc_coherent(dmadev, CQ_SIZE(depth),
934 &nvmeq->cq_dma_addr, GFP_KERNEL);
935 if (!nvmeq->cqes)
936 goto free_nvmeq;
937 memset((void *)nvmeq->cqes, 0, CQ_SIZE(depth));
938
939 nvmeq->sq_cmds = dma_alloc_coherent(dmadev, SQ_SIZE(depth),
940 &nvmeq->sq_dma_addr, GFP_KERNEL);
941 if (!nvmeq->sq_cmds)
942 goto free_cqdma;
943
944 nvmeq->q_dmadev = dmadev;
Matthew Wilcox091b6092011-02-10 09:56:01 -0500945 nvmeq->dev = dev;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500946 spin_lock_init(&nvmeq->q_lock);
947 nvmeq->cq_head = 0;
Matthew Wilcox82123462011-01-20 13:24:06 -0500948 nvmeq->cq_phase = 1;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500949 init_waitqueue_head(&nvmeq->sq_full);
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -0500950 init_waitqueue_entry(&nvmeq->sq_cong_wait, nvme_thread);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500951 bio_list_init(&nvmeq->sq_cong);
Matthew Wilcoxf1938f62011-10-20 17:00:41 -0400952 nvmeq->q_db = &dev->dbs[qid << (dev->db_stride + 1)];
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500953 nvmeq->q_depth = depth;
954 nvmeq->cq_vector = vector;
955
956 return nvmeq;
957
958 free_cqdma:
959 dma_free_coherent(dmadev, CQ_SIZE(nvmeq->q_depth), (void *)nvmeq->cqes,
960 nvmeq->cq_dma_addr);
961 free_nvmeq:
962 kfree(nvmeq);
963 return NULL;
964}
965
Matthew Wilcox30010822011-01-20 09:10:15 -0500966static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq,
967 const char *name)
968{
Matthew Wilcox58ffacb2011-02-06 07:28:06 -0500969 if (use_threaded_interrupts)
970 return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector,
Matthew Wilcoxec6ce612011-02-06 09:01:00 -0500971 nvme_irq_check, nvme_irq,
Matthew Wilcox58ffacb2011-02-06 07:28:06 -0500972 IRQF_DISABLED | IRQF_SHARED,
973 name, nvmeq);
Matthew Wilcox30010822011-01-20 09:10:15 -0500974 return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq,
975 IRQF_DISABLED | IRQF_SHARED, name, nvmeq);
976}
977
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -0800978static struct nvme_queue *nvme_create_queue(struct nvme_dev *dev, int qid,
979 int cq_size, int vector)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500980{
981 int result;
982 struct nvme_queue *nvmeq = nvme_alloc_queue(dev, qid, cq_size, vector);
983
Matthew Wilcox3f85d502011-02-01 08:39:04 -0500984 if (!nvmeq)
Matthew Wilcox6f0f5442011-05-11 13:30:59 -0700985 return ERR_PTR(-ENOMEM);
Matthew Wilcox3f85d502011-02-01 08:39:04 -0500986
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500987 result = adapter_alloc_cq(dev, qid, nvmeq);
988 if (result < 0)
989 goto free_nvmeq;
990
991 result = adapter_alloc_sq(dev, qid, nvmeq);
992 if (result < 0)
993 goto release_cq;
994
Matthew Wilcox30010822011-01-20 09:10:15 -0500995 result = queue_request_irq(dev, nvmeq, "nvme");
Matthew Wilcoxb60503b2011-01-20 12:50:14 -0500996 if (result < 0)
997 goto release_sq;
998
999 return nvmeq;
1000
1001 release_sq:
1002 adapter_delete_sq(dev, qid);
1003 release_cq:
1004 adapter_delete_cq(dev, qid);
1005 free_nvmeq:
1006 dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
1007 (void *)nvmeq->cqes, nvmeq->cq_dma_addr);
1008 dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
1009 nvmeq->sq_cmds, nvmeq->sq_dma_addr);
1010 kfree(nvmeq);
Matthew Wilcox6f0f5442011-05-11 13:30:59 -07001011 return ERR_PTR(result);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001012}
1013
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001014static int nvme_configure_admin_queue(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001015{
Matthew Wilcox9e866772012-08-03 13:55:56 -04001016 int result = 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001017 u32 aqa;
Matthew Wilcox22605f92011-04-19 15:04:20 -04001018 u64 cap;
1019 unsigned long timeout;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001020 struct nvme_queue *nvmeq;
1021
1022 dev->dbs = ((void __iomem *)dev->bar) + 4096;
1023
1024 nvmeq = nvme_alloc_queue(dev, 0, 64, 0);
Matthew Wilcox3f85d502011-02-01 08:39:04 -05001025 if (!nvmeq)
1026 return -ENOMEM;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001027
1028 aqa = nvmeq->q_depth - 1;
1029 aqa |= aqa << 16;
1030
1031 dev->ctrl_config = NVME_CC_ENABLE | NVME_CC_CSS_NVM;
1032 dev->ctrl_config |= (PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
1033 dev->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
Matthew Wilcox7f53f9d2011-03-22 15:55:45 -04001034 dev->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001035
Shane Michael Matthews5911f202011-02-01 11:31:55 -05001036 writel(0, &dev->bar->cc);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001037 writel(aqa, &dev->bar->aqa);
1038 writeq(nvmeq->sq_dma_addr, &dev->bar->asq);
1039 writeq(nvmeq->cq_dma_addr, &dev->bar->acq);
1040 writel(dev->ctrl_config, &dev->bar->cc);
1041
Matthew Wilcox22605f92011-04-19 15:04:20 -04001042 cap = readq(&dev->bar->cap);
1043 timeout = ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
Matthew Wilcoxf1938f62011-10-20 17:00:41 -04001044 dev->db_stride = NVME_CAP_STRIDE(cap);
Matthew Wilcox22605f92011-04-19 15:04:20 -04001045
Matthew Wilcox9e866772012-08-03 13:55:56 -04001046 while (!result && !(readl(&dev->bar->csts) & NVME_CSTS_RDY)) {
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001047 msleep(100);
1048 if (fatal_signal_pending(current))
Matthew Wilcox9e866772012-08-03 13:55:56 -04001049 result = -EINTR;
Matthew Wilcox22605f92011-04-19 15:04:20 -04001050 if (time_after(jiffies, timeout)) {
1051 dev_err(&dev->pci_dev->dev,
1052 "Device not ready; aborting initialisation\n");
Matthew Wilcox9e866772012-08-03 13:55:56 -04001053 result = -ENODEV;
Matthew Wilcox22605f92011-04-19 15:04:20 -04001054 }
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001055 }
1056
Matthew Wilcox9e866772012-08-03 13:55:56 -04001057 if (result) {
1058 nvme_free_queue_mem(nvmeq);
1059 return result;
1060 }
1061
Matthew Wilcox30010822011-01-20 09:10:15 -05001062 result = queue_request_irq(dev, nvmeq, "nvme admin");
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001063 dev->queues[0] = nvmeq;
1064 return result;
1065}
1066
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001067static struct nvme_iod *nvme_map_user_pages(struct nvme_dev *dev, int write,
1068 unsigned long addr, unsigned length)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001069{
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001070 int i, err, count, nents, offset;
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001071 struct scatterlist *sg;
1072 struct page **pages;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001073 struct nvme_iod *iod;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001074
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001075 if (addr & 3)
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001076 return ERR_PTR(-EINVAL);
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001077 if (!length)
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001078 return ERR_PTR(-EINVAL);
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001079
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001080 offset = offset_in_page(addr);
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001081 count = DIV_ROUND_UP(offset + length, PAGE_SIZE);
1082 pages = kcalloc(count, sizeof(*pages), GFP_KERNEL);
Dan Carpenter22fff822012-01-20 07:55:30 -05001083 if (!pages)
1084 return ERR_PTR(-ENOMEM);
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001085
1086 err = get_user_pages_fast(addr, count, 1, pages);
1087 if (err < count) {
1088 count = err;
1089 err = -EFAULT;
1090 goto put_pages;
1091 }
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001092
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001093 iod = nvme_alloc_iod(count, length, GFP_KERNEL);
1094 sg = iod->sg;
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001095 sg_init_table(sg, count);
Matthew Wilcoxd0ba1e42011-09-13 17:01:39 -04001096 for (i = 0; i < count; i++) {
1097 sg_set_page(&sg[i], pages[i],
1098 min_t(int, length, PAGE_SIZE - offset), offset);
1099 length -= (PAGE_SIZE - offset);
1100 offset = 0;
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001101 }
Matthew Wilcoxfe304c42012-01-06 13:49:25 -07001102 sg_mark_end(&sg[i - 1]);
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001103 iod->nents = count;
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001104
1105 err = -ENOMEM;
1106 nents = dma_map_sg(&dev->pci_dev->dev, sg, count,
1107 write ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001108 if (!nents)
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001109 goto free_iod;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001110
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001111 kfree(pages);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001112 return iod;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001113
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001114 free_iod:
1115 kfree(iod);
Matthew Wilcox36c14ed2011-01-24 07:52:07 -05001116 put_pages:
1117 for (i = 0; i < count; i++)
1118 put_page(pages[i]);
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001119 kfree(pages);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001120 return ERR_PTR(err);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001121}
1122
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001123static void nvme_unmap_user_pages(struct nvme_dev *dev, int write,
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001124 struct nvme_iod *iod)
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001125{
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001126 int i;
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001127
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001128 dma_unmap_sg(&dev->pci_dev->dev, iod->sg, iod->nents,
1129 write ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001130
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001131 for (i = 0; i < iod->nents; i++)
1132 put_page(sg_page(&iod->sg[i]));
Matthew Wilcox7fc3cda2011-01-26 17:05:50 -05001133}
1134
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001135static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1136{
1137 struct nvme_dev *dev = ns->dev;
1138 struct nvme_queue *nvmeq;
1139 struct nvme_user_io io;
1140 struct nvme_command c;
1141 unsigned length;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001142 int status;
1143 struct nvme_iod *iod;
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001144
1145 if (copy_from_user(&io, uio, sizeof(io)))
1146 return -EFAULT;
Matthew Wilcox6c7d4942011-03-21 09:48:57 -04001147 length = (io.nblocks + 1) << ns->lba_shift;
1148
1149 switch (io.opcode) {
1150 case nvme_cmd_write:
1151 case nvme_cmd_read:
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001152 case nvme_cmd_compare:
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001153 iod = nvme_map_user_pages(dev, io.opcode & 1, io.addr, length);
Matthew Wilcox64132142011-08-09 12:56:37 -04001154 break;
Matthew Wilcox6c7d4942011-03-21 09:48:57 -04001155 default:
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001156 return -EINVAL;
Matthew Wilcox6c7d4942011-03-21 09:48:57 -04001157 }
1158
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001159 if (IS_ERR(iod))
1160 return PTR_ERR(iod);
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001161
1162 memset(&c, 0, sizeof(c));
1163 c.rw.opcode = io.opcode;
1164 c.rw.flags = io.flags;
Matthew Wilcox6c7d4942011-03-21 09:48:57 -04001165 c.rw.nsid = cpu_to_le32(ns->ns_id);
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001166 c.rw.slba = cpu_to_le64(io.slba);
Matthew Wilcox6c7d4942011-03-21 09:48:57 -04001167 c.rw.length = cpu_to_le16(io.nblocks);
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001168 c.rw.control = cpu_to_le16(io.control);
1169 c.rw.dsmgmt = cpu_to_le16(io.dsmgmt);
Matthew Wilcox6c7d4942011-03-21 09:48:57 -04001170 c.rw.reftag = io.reftag;
1171 c.rw.apptag = io.apptag;
1172 c.rw.appmask = io.appmask;
Shane Michael Matthewse025344c2011-02-10 08:51:24 -05001173 /* XXX: metadata */
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001174 length = nvme_setup_prps(dev, &c.common, iod, length, GFP_KERNEL);
Shane Michael Matthewse025344c2011-02-10 08:51:24 -05001175
Matthew Wilcox040a93b2011-12-20 11:04:12 -05001176 nvmeq = get_nvmeq(dev);
Matthew Wilcoxfa922822011-03-16 16:29:00 -04001177 /*
1178 * Since nvme_submit_sync_cmd sleeps, we can't keep preemption
Matthew Wilcoxb1ad37e2011-02-04 16:14:30 -05001179 * disabled. We may be preempted at any point, and be rescheduled
1180 * to a different CPU. That will cause cacheline bouncing, but no
1181 * additional races since q_lock already protects against other CPUs.
1182 */
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001183 put_nvmeq(nvmeq);
Matthew Wilcoxb77954c2011-05-12 13:51:41 -04001184 if (length != (io.nblocks + 1) << ns->lba_shift)
1185 status = -ENOMEM;
1186 else
Matthew Wilcoxff976d72011-12-20 13:53:01 -05001187 status = nvme_submit_sync_cmd(nvmeq, &c, NULL, NVME_IO_TIMEOUT);
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001188
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001189 nvme_unmap_user_pages(dev, io.opcode & 1, iod);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001190 nvme_free_iod(dev, iod);
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001191 return status;
1192}
1193
Keith Busch50af8ba2012-07-25 16:07:55 -06001194static int nvme_user_admin_cmd(struct nvme_dev *dev,
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001195 struct nvme_admin_cmd __user *ucmd)
Matthew Wilcox6ee44cd2011-02-03 10:58:26 -05001196{
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001197 struct nvme_admin_cmd cmd;
Matthew Wilcox6ee44cd2011-02-03 10:58:26 -05001198 struct nvme_command c;
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001199 int status, length;
Keith Buschc7d36ab2012-07-27 11:53:28 -06001200 struct nvme_iod *uninitialized_var(iod);
Matthew Wilcox6ee44cd2011-02-03 10:58:26 -05001201
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001202 if (!capable(CAP_SYS_ADMIN))
1203 return -EACCES;
1204 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
Matthew Wilcox6ee44cd2011-02-03 10:58:26 -05001205 return -EFAULT;
Matthew Wilcox6ee44cd2011-02-03 10:58:26 -05001206
1207 memset(&c, 0, sizeof(c));
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001208 c.common.opcode = cmd.opcode;
1209 c.common.flags = cmd.flags;
1210 c.common.nsid = cpu_to_le32(cmd.nsid);
1211 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1212 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1213 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1214 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1215 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1216 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1217 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1218 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1219
1220 length = cmd.data_len;
1221 if (cmd.data_len) {
Matthew Wilcox49742182012-01-06 13:42:45 -07001222 iod = nvme_map_user_pages(dev, cmd.opcode & 1, cmd.addr,
1223 length);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001224 if (IS_ERR(iod))
1225 return PTR_ERR(iod);
1226 length = nvme_setup_prps(dev, &c.common, iod, length,
1227 GFP_KERNEL);
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001228 }
1229
1230 if (length != cmd.data_len)
Matthew Wilcoxb77954c2011-05-12 13:51:41 -04001231 status = -ENOMEM;
1232 else
1233 status = nvme_submit_admin_cmd(dev, &c, NULL);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001234
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001235 if (cmd.data_len) {
Matthew Wilcox1c2ad9f2012-01-06 13:52:56 -07001236 nvme_unmap_user_pages(dev, cmd.opcode & 1, iod);
Matthew Wilcoxeca18b22011-12-20 13:34:52 -05001237 nvme_free_iod(dev, iod);
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001238 }
Matthew Wilcox6ee44cd2011-02-03 10:58:26 -05001239 return status;
1240}
1241
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001242static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd,
1243 unsigned long arg)
1244{
1245 struct nvme_ns *ns = bdev->bd_disk->private_data;
1246
1247 switch (cmd) {
Matthew Wilcox6bbf1ac2011-05-20 13:03:42 -04001248 case NVME_IOCTL_ID:
1249 return ns->ns_id;
1250 case NVME_IOCTL_ADMIN_CMD:
Keith Busch50af8ba2012-07-25 16:07:55 -06001251 return nvme_user_admin_cmd(ns->dev, (void __user *)arg);
Matthew Wilcoxa53295b2011-02-01 16:13:29 -05001252 case NVME_IOCTL_SUBMIT_IO:
1253 return nvme_submit_io(ns, (void __user *)arg);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001254 default:
1255 return -ENOTTY;
1256 }
1257}
1258
1259static const struct block_device_operations nvme_fops = {
1260 .owner = THIS_MODULE,
1261 .ioctl = nvme_ioctl,
Matthew Wilcox49481682011-03-19 14:55:38 -04001262 .compat_ioctl = nvme_ioctl,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001263};
1264
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001265static void nvme_resubmit_bios(struct nvme_queue *nvmeq)
1266{
1267 while (bio_list_peek(&nvmeq->sq_cong)) {
1268 struct bio *bio = bio_list_pop(&nvmeq->sq_cong);
1269 struct nvme_ns *ns = bio->bi_bdev->bd_disk->private_data;
1270 if (nvme_submit_bio_queue(nvmeq, ns, bio)) {
1271 bio_list_add_head(&nvmeq->sq_cong, bio);
1272 break;
1273 }
Matthew Wilcox3cb967c2011-03-16 16:45:49 -04001274 if (bio_list_empty(&nvmeq->sq_cong))
1275 remove_wait_queue(&nvmeq->sq_full,
1276 &nvmeq->sq_cong_wait);
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001277 }
1278}
1279
1280static int nvme_kthread(void *data)
1281{
1282 struct nvme_dev *dev;
1283
1284 while (!kthread_should_stop()) {
1285 __set_current_state(TASK_RUNNING);
1286 spin_lock(&dev_list_lock);
1287 list_for_each_entry(dev, &dev_list, node) {
1288 int i;
1289 for (i = 0; i < dev->queue_count; i++) {
1290 struct nvme_queue *nvmeq = dev->queues[i];
Matthew Wilcox740216f2011-02-15 16:28:20 -05001291 if (!nvmeq)
1292 continue;
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001293 spin_lock_irq(&nvmeq->q_lock);
1294 if (nvme_process_cq(nvmeq))
1295 printk("process_cq did something\n");
Matthew Wilcoxa09115b2012-08-07 15:56:23 -04001296 nvme_cancel_ios(nvmeq, true);
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001297 nvme_resubmit_bios(nvmeq);
1298 spin_unlock_irq(&nvmeq->q_lock);
1299 }
1300 }
1301 spin_unlock(&dev_list_lock);
1302 set_current_state(TASK_INTERRUPTIBLE);
1303 schedule_timeout(HZ);
1304 }
1305 return 0;
1306}
1307
Matthew Wilcox5aff9382011-05-06 08:45:47 -04001308static DEFINE_IDA(nvme_index_ida);
1309
1310static int nvme_get_ns_idx(void)
1311{
1312 int index, error;
1313
1314 do {
1315 if (!ida_pre_get(&nvme_index_ida, GFP_KERNEL))
1316 return -1;
1317
1318 spin_lock(&dev_list_lock);
1319 error = ida_get_new(&nvme_index_ida, &index);
1320 spin_unlock(&dev_list_lock);
1321 } while (error == -EAGAIN);
1322
1323 if (error)
1324 index = -1;
1325 return index;
1326}
1327
1328static void nvme_put_ns_idx(int index)
1329{
1330 spin_lock(&dev_list_lock);
1331 ida_remove(&nvme_index_ida, index);
1332 spin_unlock(&dev_list_lock);
1333}
1334
1335static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, int nsid,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001336 struct nvme_id_ns *id, struct nvme_lba_range_type *rt)
1337{
1338 struct nvme_ns *ns;
1339 struct gendisk *disk;
1340 int lbaf;
1341
1342 if (rt->attributes & NVME_LBART_ATTRIB_HIDE)
1343 return NULL;
1344
1345 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1346 if (!ns)
1347 return NULL;
1348 ns->queue = blk_alloc_queue(GFP_KERNEL);
1349 if (!ns->queue)
1350 goto out_free_ns;
Matthew Wilcox4eeb9212012-01-10 14:35:08 -07001351 ns->queue->queue_flags = QUEUE_FLAG_DEFAULT;
1352 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
1353 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1354/* queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); */
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001355 blk_queue_make_request(ns->queue, nvme_make_request);
1356 ns->dev = dev;
1357 ns->queue->queuedata = ns;
1358
1359 disk = alloc_disk(NVME_MINORS);
1360 if (!disk)
1361 goto out_free_queue;
Matthew Wilcox5aff9382011-05-06 08:45:47 -04001362 ns->ns_id = nsid;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001363 ns->disk = disk;
1364 lbaf = id->flbas & 0xf;
1365 ns->lba_shift = id->lbaf[lbaf].ds;
Keith Busche9ef4632012-07-24 15:01:04 -06001366 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
Keith Busch8fc23e02012-07-26 11:29:57 -06001367 if (dev->max_hw_sectors)
1368 blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001369
1370 disk->major = nvme_major;
1371 disk->minors = NVME_MINORS;
Matthew Wilcox5aff9382011-05-06 08:45:47 -04001372 disk->first_minor = NVME_MINORS * nvme_get_ns_idx();
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001373 disk->fops = &nvme_fops;
1374 disk->private_data = ns;
1375 disk->queue = ns->queue;
Matthew Wilcox388f0372011-02-01 12:49:38 -05001376 disk->driverfs_dev = &dev->pci_dev->dev;
Matthew Wilcox5aff9382011-05-06 08:45:47 -04001377 sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001378 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1379
1380 return ns;
1381
1382 out_free_queue:
1383 blk_cleanup_queue(ns->queue);
1384 out_free_ns:
1385 kfree(ns);
1386 return NULL;
1387}
1388
1389static void nvme_ns_free(struct nvme_ns *ns)
1390{
Matthew Wilcox5aff9382011-05-06 08:45:47 -04001391 int index = ns->disk->first_minor / NVME_MINORS;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001392 put_disk(ns->disk);
Matthew Wilcox5aff9382011-05-06 08:45:47 -04001393 nvme_put_ns_idx(index);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001394 blk_cleanup_queue(ns->queue);
1395 kfree(ns);
1396}
1397
Matthew Wilcoxb3b06812011-01-20 09:14:34 -05001398static int set_queue_count(struct nvme_dev *dev, int count)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001399{
1400 int status;
1401 u32 result;
Matthew Wilcoxb3b06812011-01-20 09:14:34 -05001402 u32 q_count = (count - 1) | ((count - 1) << 16);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001403
Matthew Wilcoxdf348132012-01-11 07:29:56 -07001404 status = nvme_set_features(dev, NVME_FEAT_NUM_QUEUES, q_count, 0,
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001405 &result);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001406 if (status)
1407 return -EIO;
1408 return min(result & 0xffff, result >> 16) + 1;
1409}
1410
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001411static int nvme_setup_io_queues(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001412{
Keith Buscha0cadb82012-07-27 13:57:23 -04001413 int result, cpu, i, nr_io_queues, db_bar_size, q_depth;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001414
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001415 nr_io_queues = num_online_cpus();
1416 result = set_queue_count(dev, nr_io_queues);
Matthew Wilcox1b234842011-01-20 13:01:49 -05001417 if (result < 0)
1418 return result;
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001419 if (result < nr_io_queues)
1420 nr_io_queues = result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001421
Matthew Wilcox1b234842011-01-20 13:01:49 -05001422 /* Deregister the admin queue's interrupt */
1423 free_irq(dev->entry[0].vector, dev->queues[0]);
1424
Matthew Wilcoxf1938f62011-10-20 17:00:41 -04001425 db_bar_size = 4096 + ((nr_io_queues + 1) << (dev->db_stride + 3));
1426 if (db_bar_size > 8192) {
1427 iounmap(dev->bar);
1428 dev->bar = ioremap(pci_resource_start(dev->pci_dev, 0),
1429 db_bar_size);
1430 dev->dbs = ((void __iomem *)dev->bar) + 4096;
1431 dev->queues[0]->q_db = dev->dbs;
1432 }
1433
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001434 for (i = 0; i < nr_io_queues; i++)
Matthew Wilcox1b234842011-01-20 13:01:49 -05001435 dev->entry[i].entry = i;
1436 for (;;) {
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001437 result = pci_enable_msix(dev->pci_dev, dev->entry,
1438 nr_io_queues);
Matthew Wilcox1b234842011-01-20 13:01:49 -05001439 if (result == 0) {
1440 break;
1441 } else if (result > 0) {
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001442 nr_io_queues = result;
Matthew Wilcox1b234842011-01-20 13:01:49 -05001443 continue;
1444 } else {
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001445 nr_io_queues = 1;
Matthew Wilcox1b234842011-01-20 13:01:49 -05001446 break;
1447 }
1448 }
1449
1450 result = queue_request_irq(dev, dev->queues[0], "nvme admin");
1451 /* XXX: handle failure here */
1452
1453 cpu = cpumask_first(cpu_online_mask);
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001454 for (i = 0; i < nr_io_queues; i++) {
Matthew Wilcox1b234842011-01-20 13:01:49 -05001455 irq_set_affinity_hint(dev->entry[i].vector, get_cpu_mask(cpu));
1456 cpu = cpumask_next(cpu, cpu_online_mask);
1457 }
1458
Keith Buscha0cadb82012-07-27 13:57:23 -04001459 q_depth = min_t(int, NVME_CAP_MQES(readq(&dev->bar->cap)) + 1,
1460 NVME_Q_DEPTH);
Matthew Wilcoxb348b7d2011-02-15 16:16:02 -05001461 for (i = 0; i < nr_io_queues; i++) {
Keith Buscha0cadb82012-07-27 13:57:23 -04001462 dev->queues[i + 1] = nvme_create_queue(dev, i + 1, q_depth, i);
Matthew Wilcox6f0f5442011-05-11 13:30:59 -07001463 if (IS_ERR(dev->queues[i + 1]))
1464 return PTR_ERR(dev->queues[i + 1]);
Matthew Wilcox1b234842011-01-20 13:01:49 -05001465 dev->queue_count++;
1466 }
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001467
Matthew Wilcox9ecdc942011-03-16 16:52:19 -04001468 for (; i < num_possible_cpus(); i++) {
1469 int target = i % rounddown_pow_of_two(dev->queue_count - 1);
1470 dev->queues[i + 1] = dev->queues[target + 1];
1471 }
1472
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001473 return 0;
1474}
1475
1476static void nvme_free_queues(struct nvme_dev *dev)
1477{
1478 int i;
1479
1480 for (i = dev->queue_count - 1; i >= 0; i--)
1481 nvme_free_queue(dev, i);
1482}
1483
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001484static int nvme_dev_add(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001485{
1486 int res, nn, i;
1487 struct nvme_ns *ns, *next;
Matthew Wilcox51814232011-02-01 16:18:08 -05001488 struct nvme_id_ctrl *ctrl;
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001489 struct nvme_id_ns *id_ns;
1490 void *mem;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001491 dma_addr_t dma_addr;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001492
1493 res = nvme_setup_io_queues(dev);
1494 if (res)
1495 return res;
1496
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001497 mem = dma_alloc_coherent(&dev->pci_dev->dev, 8192, &dma_addr,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001498 GFP_KERNEL);
1499
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001500 res = nvme_identify(dev, 0, 1, dma_addr);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001501 if (res) {
1502 res = -EIO;
1503 goto out_free;
1504 }
1505
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001506 ctrl = mem;
Matthew Wilcox51814232011-02-01 16:18:08 -05001507 nn = le32_to_cpup(&ctrl->nn);
1508 memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn));
1509 memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn));
1510 memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
Keith Busch8fc23e02012-07-26 11:29:57 -06001511 if (ctrl->mdts) {
1512 int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12;
1513 dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9);
1514 }
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001515
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001516 id_ns = mem;
Matthew Wilcox2b2c1892011-10-07 13:10:13 -04001517 for (i = 1; i <= nn; i++) {
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001518 res = nvme_identify(dev, i, 0, dma_addr);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001519 if (res)
1520 continue;
1521
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001522 if (id_ns->ncap == 0)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001523 continue;
1524
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001525 res = nvme_get_features(dev, NVME_FEAT_LBA_RANGE, i,
Matthew Wilcoxdf348132012-01-11 07:29:56 -07001526 dma_addr + 4096);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001527 if (res)
1528 continue;
1529
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001530 ns = nvme_alloc_ns(dev, i, mem, mem + 4096);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001531 if (ns)
1532 list_add_tail(&ns->list, &dev->namespaces);
1533 }
1534 list_for_each_entry(ns, &dev->namespaces, list)
1535 add_disk(ns->disk);
1536
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001537 goto out;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001538
1539 out_free:
1540 list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
1541 list_del(&ns->list);
1542 nvme_ns_free(ns);
1543 }
1544
Matthew Wilcoxbc5fc7e2011-09-19 17:08:14 -04001545 out:
Matthew Wilcox684f5c22011-09-19 17:14:53 -04001546 dma_free_coherent(&dev->pci_dev->dev, 8192, mem, dma_addr);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001547 return res;
1548}
1549
1550static int nvme_dev_remove(struct nvme_dev *dev)
1551{
1552 struct nvme_ns *ns, *next;
1553
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001554 spin_lock(&dev_list_lock);
1555 list_del(&dev->node);
1556 spin_unlock(&dev_list_lock);
1557
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001558 list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
1559 list_del(&ns->list);
1560 del_gendisk(ns->disk);
1561 nvme_ns_free(ns);
1562 }
1563
1564 nvme_free_queues(dev);
1565
1566 return 0;
1567}
1568
Matthew Wilcox091b6092011-02-10 09:56:01 -05001569static int nvme_setup_prp_pools(struct nvme_dev *dev)
1570{
1571 struct device *dmadev = &dev->pci_dev->dev;
1572 dev->prp_page_pool = dma_pool_create("prp list page", dmadev,
1573 PAGE_SIZE, PAGE_SIZE, 0);
1574 if (!dev->prp_page_pool)
1575 return -ENOMEM;
1576
Matthew Wilcox99802a72011-02-10 10:30:34 -05001577 /* Optimisation for I/Os between 4k and 128k */
1578 dev->prp_small_pool = dma_pool_create("prp list 256", dmadev,
1579 256, 256, 0);
1580 if (!dev->prp_small_pool) {
1581 dma_pool_destroy(dev->prp_page_pool);
1582 return -ENOMEM;
1583 }
Matthew Wilcox091b6092011-02-10 09:56:01 -05001584 return 0;
1585}
1586
1587static void nvme_release_prp_pools(struct nvme_dev *dev)
1588{
1589 dma_pool_destroy(dev->prp_page_pool);
Matthew Wilcox99802a72011-02-10 10:30:34 -05001590 dma_pool_destroy(dev->prp_small_pool);
Matthew Wilcox091b6092011-02-10 09:56:01 -05001591}
1592
Quoc-Son Anhcd58ad72012-02-21 16:50:53 -07001593static DEFINE_IDA(nvme_instance_ida);
1594
1595static int nvme_set_instance(struct nvme_dev *dev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001596{
Quoc-Son Anhcd58ad72012-02-21 16:50:53 -07001597 int instance, error;
1598
1599 do {
1600 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1601 return -ENODEV;
1602
1603 spin_lock(&dev_list_lock);
1604 error = ida_get_new(&nvme_instance_ida, &instance);
1605 spin_unlock(&dev_list_lock);
1606 } while (error == -EAGAIN);
1607
1608 if (error)
1609 return -ENODEV;
1610
1611 dev->instance = instance;
1612 return 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001613}
1614
1615static void nvme_release_instance(struct nvme_dev *dev)
1616{
Quoc-Son Anhcd58ad72012-02-21 16:50:53 -07001617 spin_lock(&dev_list_lock);
1618 ida_remove(&nvme_instance_ida, dev->instance);
1619 spin_unlock(&dev_list_lock);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001620}
1621
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001622static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001623{
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001624 int bars, result = -ENOMEM;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001625 struct nvme_dev *dev;
1626
1627 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1628 if (!dev)
1629 return -ENOMEM;
1630 dev->entry = kcalloc(num_possible_cpus(), sizeof(*dev->entry),
1631 GFP_KERNEL);
1632 if (!dev->entry)
1633 goto free;
Matthew Wilcox1b234842011-01-20 13:01:49 -05001634 dev->queues = kcalloc(num_possible_cpus() + 1, sizeof(void *),
1635 GFP_KERNEL);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001636 if (!dev->queues)
1637 goto free;
1638
Shane Michael Matthews0ee5a7d2011-02-01 08:49:30 -05001639 if (pci_enable_device_mem(pdev))
1640 goto free;
Matthew Wilcoxf64d3362011-02-01 09:01:59 -05001641 pci_set_master(pdev);
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001642 bars = pci_select_bars(pdev, IORESOURCE_MEM);
1643 if (pci_request_selected_regions(pdev, bars, "nvme"))
1644 goto disable;
Shane Michael Matthews0ee5a7d2011-02-01 08:49:30 -05001645
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001646 INIT_LIST_HEAD(&dev->namespaces);
1647 dev->pci_dev = pdev;
1648 pci_set_drvdata(pdev, dev);
Matthew Wilcox29303532011-02-01 16:23:39 -05001649 dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1650 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
Quoc-Son Anhcd58ad72012-02-21 16:50:53 -07001651 result = nvme_set_instance(dev);
1652 if (result)
1653 goto disable;
1654
Matthew Wilcox53c95772011-01-20 13:42:34 -05001655 dev->entry[0].vector = pdev->irq;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001656
Matthew Wilcox091b6092011-02-10 09:56:01 -05001657 result = nvme_setup_prp_pools(dev);
1658 if (result)
1659 goto disable_msix;
1660
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001661 dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
1662 if (!dev->bar) {
1663 result = -ENOMEM;
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001664 goto disable_msix;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001665 }
1666
1667 result = nvme_configure_admin_queue(dev);
1668 if (result)
1669 goto unmap;
1670 dev->queue_count++;
1671
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001672 spin_lock(&dev_list_lock);
1673 list_add(&dev->node, &dev_list);
1674 spin_unlock(&dev_list_lock);
1675
Matthew Wilcox740216f2011-02-15 16:28:20 -05001676 result = nvme_dev_add(dev);
1677 if (result)
1678 goto delete;
1679
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001680 return 0;
1681
1682 delete:
Matthew Wilcox740216f2011-02-15 16:28:20 -05001683 spin_lock(&dev_list_lock);
1684 list_del(&dev->node);
1685 spin_unlock(&dev_list_lock);
1686
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001687 nvme_free_queues(dev);
1688 unmap:
1689 iounmap(dev->bar);
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001690 disable_msix:
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001691 pci_disable_msix(pdev);
1692 nvme_release_instance(dev);
Matthew Wilcox091b6092011-02-10 09:56:01 -05001693 nvme_release_prp_pools(dev);
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001694 disable:
Shane Michael Matthews0ee5a7d2011-02-01 08:49:30 -05001695 pci_disable_device(pdev);
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001696 pci_release_regions(pdev);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001697 free:
1698 kfree(dev->queues);
1699 kfree(dev->entry);
1700 kfree(dev);
1701 return result;
1702}
1703
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001704static void nvme_remove(struct pci_dev *pdev)
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001705{
1706 struct nvme_dev *dev = pci_get_drvdata(pdev);
1707 nvme_dev_remove(dev);
1708 pci_disable_msix(pdev);
1709 iounmap(dev->bar);
1710 nvme_release_instance(dev);
Matthew Wilcox091b6092011-02-10 09:56:01 -05001711 nvme_release_prp_pools(dev);
Shane Michael Matthews0ee5a7d2011-02-01 08:49:30 -05001712 pci_disable_device(pdev);
Matthew Wilcox574e8b92011-02-01 16:24:35 -05001713 pci_release_regions(pdev);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001714 kfree(dev->queues);
1715 kfree(dev->entry);
1716 kfree(dev);
1717}
1718
1719/* These functions are yet to be implemented */
1720#define nvme_error_detected NULL
1721#define nvme_dump_registers NULL
1722#define nvme_link_reset NULL
1723#define nvme_slot_reset NULL
1724#define nvme_error_resume NULL
1725#define nvme_suspend NULL
1726#define nvme_resume NULL
1727
Stephen Hemminger1d352032012-09-07 09:33:17 -07001728static const struct pci_error_handlers nvme_err_handler = {
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001729 .error_detected = nvme_error_detected,
1730 .mmio_enabled = nvme_dump_registers,
1731 .link_reset = nvme_link_reset,
1732 .slot_reset = nvme_slot_reset,
1733 .resume = nvme_error_resume,
1734};
1735
1736/* Move to pci_ids.h later */
1737#define PCI_CLASS_STORAGE_EXPRESS 0x010802
1738
1739static DEFINE_PCI_DEVICE_TABLE(nvme_id_table) = {
1740 { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
1741 { 0, }
1742};
1743MODULE_DEVICE_TABLE(pci, nvme_id_table);
1744
1745static struct pci_driver nvme_driver = {
1746 .name = "nvme",
1747 .id_table = nvme_id_table,
1748 .probe = nvme_probe,
Greg Kroah-Hartman8d85fce2012-12-21 15:13:49 -08001749 .remove = nvme_remove,
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001750 .suspend = nvme_suspend,
1751 .resume = nvme_resume,
1752 .err_handler = &nvme_err_handler,
1753};
1754
1755static int __init nvme_init(void)
1756{
Matthew Wilcox0ac13142012-07-31 13:31:15 -04001757 int result;
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001758
1759 nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
1760 if (IS_ERR(nvme_thread))
1761 return PTR_ERR(nvme_thread);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001762
Keith Busch5c42ea12012-07-25 16:05:18 -06001763 result = register_blkdev(nvme_major, "nvme");
1764 if (result < 0)
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001765 goto kill_kthread;
Keith Busch5c42ea12012-07-25 16:05:18 -06001766 else if (result > 0)
Matthew Wilcox0ac13142012-07-31 13:31:15 -04001767 nvme_major = result;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001768
1769 result = pci_register_driver(&nvme_driver);
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001770 if (result)
1771 goto unregister_blkdev;
1772 return 0;
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001773
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001774 unregister_blkdev:
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001775 unregister_blkdev(nvme_major, "nvme");
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001776 kill_kthread:
1777 kthread_stop(nvme_thread);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001778 return result;
1779}
1780
1781static void __exit nvme_exit(void)
1782{
1783 pci_unregister_driver(&nvme_driver);
1784 unregister_blkdev(nvme_major, "nvme");
Matthew Wilcox1fa6aea2011-03-02 18:37:18 -05001785 kthread_stop(nvme_thread);
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001786}
1787
1788MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
1789MODULE_LICENSE("GPL");
Matthew Wilcox366e8212012-01-10 16:30:15 -05001790MODULE_VERSION("0.8");
Matthew Wilcoxb60503b2011-01-20 12:50:14 -05001791module_init(nvme_init);
1792module_exit(nvme_exit);